# Copyright (c) Facebook, Inc. and its affiliates. from typing import Tuple import torch from torch import nn from torch.nn import functional as F from detectron2.config import configurable from detectron2.data import MetadataCatalog from detectron2.modeling import META_ARCH_REGISTRY, build_backbone, build_sem_seg_head from detectron2.modeling.backbone import Backbone from detectron2.modeling.postprocessing import sem_seg_postprocess from detectron2.structures import Boxes, ImageList, Instances, BitMasks from detectron2.utils.memory import retry_if_cuda_oom from .modeling.criterion import SetCriterion from .modeling.matcher import HungarianMatcher from .modeling.transformer_decoder.fcclip_transformer_decoder import MaskPooling, get_classification_logits import os VILD_PROMPT = [ "a photo of a {}.", "This is a photo of a {}", "There is a {} in the scene", "There is the {} in the scene", "a photo of a {} in the scene", "a photo of a small {}.", "a photo of a medium {}.", "a photo of a large {}.", "This is a photo of a small {}.", "This is a photo of a medium {}.", "This is a photo of a large {}.", "There is a small {} in the scene.", "There is a medium {} in the scene.", "There is a large {} in the scene.", ] def split_labels(x): res = [] for x_ in x: x_ = x_.replace(', ', ',') x_ = x_.split(',') # there can be multiple synonyms for single class res.append(x_) return res def fill_all_templates_ensemble(x_=''): res = [] for x in x_: for template in VILD_PROMPT: res.append(template.format(x)) return res, len(res) // len(VILD_PROMPT) @META_ARCH_REGISTRY.register() class FCCLIP(nn.Module): """ Main class for mask classification semantic segmentation architectures. """ @configurable def __init__( self, *, backbone: Backbone, sem_seg_head: nn.Module, criterion: nn.Module, num_queries: int, object_mask_threshold: float, overlap_threshold: float, train_metadata, test_metadata, size_divisibility: int, sem_seg_postprocess_before_inference: bool, pixel_mean: Tuple[float], pixel_std: Tuple[float], # inference semantic_on: bool, panoptic_on: bool, instance_on: bool, test_topk_per_image: int, # FC-CLIP geometric_ensemble_alpha: float, geometric_ensemble_beta: float, ): """ Args: backbone: a backbone module, must follow detectron2's backbone interface sem_seg_head: a module that predicts semantic segmentation from backbone features criterion: a module that defines the loss num_queries: int, number of queries object_mask_threshold: float, threshold to filter query based on classification score for panoptic segmentation inference overlap_threshold: overlap threshold used in general inference for panoptic segmentation metadata: dataset meta, get `thing` and `stuff` category names for panoptic segmentation inference size_divisibility: Some backbones require the input height and width to be divisible by a specific integer. We can use this to override such requirement. sem_seg_postprocess_before_inference: whether to resize the prediction back to original input size before semantic segmentation inference or after. For high-resolution dataset like Mapillary, resizing predictions before inference will cause OOM error. pixel_mean, pixel_std: list or tuple with #channels element, representing the per-channel mean and std to be used to normalize the input image semantic_on: bool, whether to output semantic segmentation prediction instance_on: bool, whether to output instance segmentation prediction panoptic_on: bool, whether to output panoptic segmentation prediction test_topk_per_image: int, instance segmentation parameter, keep topk instances per image """ super().__init__() self.backbone = backbone self.sem_seg_head = sem_seg_head self.criterion = criterion self.num_queries = num_queries self.overlap_threshold = overlap_threshold self.object_mask_threshold = object_mask_threshold self.train_metadata = train_metadata self.test_metadata = test_metadata if size_divisibility < 0: # use backbone size_divisibility if not set size_divisibility = self.backbone.size_divisibility self.size_divisibility = size_divisibility self.sem_seg_postprocess_before_inference = sem_seg_postprocess_before_inference self.register_buffer("pixel_mean", torch.Tensor(pixel_mean).view(-1, 1, 1), False) self.register_buffer("pixel_std", torch.Tensor(pixel_std).view(-1, 1, 1), False) # additional args self.semantic_on = semantic_on self.instance_on = instance_on self.panoptic_on = panoptic_on self.test_topk_per_image = test_topk_per_image if not self.semantic_on: assert self.sem_seg_postprocess_before_inference # FC-CLIP args self.mask_pooling = MaskPooling() self.geometric_ensemble_alpha = geometric_ensemble_alpha self.geometric_ensemble_beta = geometric_ensemble_beta self.train_text_classifier = None self.test_text_classifier = None self.void_embedding = nn.Embedding(1, backbone.dim_latent) # use this for void _, self.train_num_templates, self.train_class_names = self.prepare_class_names_from_metadata(train_metadata, train_metadata) self.category_overlapping_mask, self.test_num_templates, self.test_class_names = self.prepare_class_names_from_metadata(test_metadata, train_metadata) self.demo_all_text_embedding_cache = {} # This consists of COCO, ADE20K, LVIS if os.path.exists("demo_all_text_embedding_cache.pth"): # key: str of class name, value: tensor in shape of C self.demo_all_text_embedding_cache = torch.load("demo_all_text_embedding_cache.pth", map_location=self.device) self.demo_all_text_embedding_cache = {k:v.to(self.device) for k,v in self.demo_all_text_embedding_cache.items()} def prepare_class_names_from_metadata(self, metadata, train_metadata): # get text classifier try: class_names = split_labels(metadata.stuff_classes) # it includes both thing and stuff train_class_names = split_labels(train_metadata.stuff_classes) except: # this could be for insseg, where only thing_classes are available class_names = split_labels(metadata.thing_classes) train_class_names = split_labels(train_metadata.thing_classes) train_class_names = {l for label in train_class_names for l in label} category_overlapping_list = [] for test_class_names in class_names: is_overlapping = not set(train_class_names).isdisjoint(set(test_class_names)) category_overlapping_list.append(is_overlapping) category_overlapping_mask = torch.tensor( category_overlapping_list, dtype=torch.long) num_templates = [] templated_class_names = [] for x in class_names: templated_classes, templated_classes_num = fill_all_templates_ensemble(x) templated_class_names += templated_classes num_templates.append(templated_classes_num) # how many templates for current classes class_names = templated_class_names #print("text for classification:", class_names) return category_overlapping_mask, num_templates, class_names def set_metadata(self, metadata): if set(self.test_metadata.stuff_classes) != set(metadata.stuff_classes): print("setting test metadata:", metadata) self.test_metadata = metadata self.category_overlapping_mask, self.test_num_templates, self.test_class_names = self.prepare_class_names_from_metadata(metadata, self.train_metadata) self.test_text_classifier = None print("text for classification:", self.test_class_names) return def get_text_classifier(self): if self.training: if self.train_text_classifier is None: text_classifier = [] # this is needed to avoid oom, which may happen when num of class is large bs = 128 for idx in range(0, len(self.train_class_names), bs): text_classifier.append(self.backbone.get_text_classifier(self.train_class_names[idx:idx+bs], self.device).detach()) text_classifier = torch.cat(text_classifier, dim=0) # average across templates and normalization. text_classifier /= text_classifier.norm(dim=-1, keepdim=True) text_classifier = text_classifier.reshape(text_classifier.shape[0]//len(VILD_PROMPT), len(VILD_PROMPT), text_classifier.shape[-1]).mean(1) text_classifier /= text_classifier.norm(dim=-1, keepdim=True) self.train_text_classifier = text_classifier return self.train_text_classifier, self.train_num_templates else: if self.test_text_classifier is None: try: nontemplated_class_names = split_labels(self.test_metadata.stuff_classes) # it includes both thing and stuff except: # this could be for insseg, where only thing_classes are available nontemplated_class_names = split_labels(self.test_metadata.thing_classes) print("nontemplated_class_names:", nontemplated_class_names) text2classifier = {} test_class_names = [] uncached_class_name = [] text_classifier = [] # exclude those already in cache for class_names in nontemplated_class_names: if not isinstance(class_names, list): class_names = [class_names] for class_name in class_names: if class_name in self.demo_all_text_embedding_cache: text2classifier[class_name] = self.demo_all_text_embedding_cache[class_name].to(self.device) else: test_class_names += fill_all_templates_ensemble([class_name])[0] uncached_class_name.append(class_name) print("Uncached texts:", len(uncached_class_name), uncached_class_name, test_class_names) # this is needed to avoid oom, which may happen when num of class is large bs = 128 for idx in range(0, len(test_class_names), bs): text_classifier.append(self.backbone.get_text_classifier(test_class_names[idx:idx+bs], self.device).detach()) if len(text_classifier) > 0: text_classifier = torch.cat(text_classifier, dim=0) # average across templates and normalization. text_classifier /= text_classifier.norm(dim=-1, keepdim=True) text_classifier = text_classifier.reshape(text_classifier.shape[0]//len(VILD_PROMPT), len(VILD_PROMPT), text_classifier.shape[-1]).mean(1) text_classifier /= text_classifier.norm(dim=-1, keepdim=True) assert text_classifier.shape[0] == len(uncached_class_name) for idx in range(len(uncached_class_name)): self.demo_all_text_embedding_cache[uncached_class_name[idx]] = text_classifier[idx] text2classifier[uncached_class_name[idx]] = text_classifier[idx] #torch.save({k:v for k, v in self.demo_all_text_embedding_cache.items()}, "demo_all_text_embedding_cache.pth") text_classifier = [] for class_names in nontemplated_class_names: for text in class_names: text_classifier.append(text2classifier[text].to(self.device)) text_classifier = torch.stack(text_classifier, dim=0).to(self.device) self.test_text_classifier = text_classifier return self.test_text_classifier, self.test_num_templates @classmethod def from_config(cls, cfg): backbone = build_backbone(cfg) sem_seg_head = build_sem_seg_head(cfg, backbone.output_shape()) # Loss parameters: deep_supervision = cfg.MODEL.MASK_FORMER.DEEP_SUPERVISION no_object_weight = cfg.MODEL.MASK_FORMER.NO_OBJECT_WEIGHT # loss weights class_weight = cfg.MODEL.MASK_FORMER.CLASS_WEIGHT dice_weight = cfg.MODEL.MASK_FORMER.DICE_WEIGHT mask_weight = cfg.MODEL.MASK_FORMER.MASK_WEIGHT # building criterion matcher = HungarianMatcher( cost_class=class_weight, cost_mask=mask_weight, cost_dice=dice_weight, num_points=cfg.MODEL.MASK_FORMER.TRAIN_NUM_POINTS, ) weight_dict = {"loss_ce": class_weight, "loss_mask": mask_weight, "loss_dice": dice_weight} if deep_supervision: dec_layers = cfg.MODEL.MASK_FORMER.DEC_LAYERS aux_weight_dict = {} for i in range(dec_layers - 1): aux_weight_dict.update({k + f"_{i}": v for k, v in weight_dict.items()}) weight_dict.update(aux_weight_dict) losses = ["labels", "masks"] criterion = SetCriterion( sem_seg_head.num_classes, matcher=matcher, weight_dict=weight_dict, eos_coef=no_object_weight, losses=losses, num_points=cfg.MODEL.MASK_FORMER.TRAIN_NUM_POINTS, oversample_ratio=cfg.MODEL.MASK_FORMER.OVERSAMPLE_RATIO, importance_sample_ratio=cfg.MODEL.MASK_FORMER.IMPORTANCE_SAMPLE_RATIO, ) return { "backbone": backbone, "sem_seg_head": sem_seg_head, "criterion": criterion, "num_queries": cfg.MODEL.MASK_FORMER.NUM_OBJECT_QUERIES, "object_mask_threshold": cfg.MODEL.MASK_FORMER.TEST.OBJECT_MASK_THRESHOLD, "overlap_threshold": cfg.MODEL.MASK_FORMER.TEST.OVERLAP_THRESHOLD, "train_metadata": MetadataCatalog.get(cfg.DATASETS.TRAIN[0]), "test_metadata": MetadataCatalog.get(cfg.DATASETS.TEST[0]), "size_divisibility": cfg.MODEL.MASK_FORMER.SIZE_DIVISIBILITY, "sem_seg_postprocess_before_inference": ( cfg.MODEL.MASK_FORMER.TEST.SEM_SEG_POSTPROCESSING_BEFORE_INFERENCE or cfg.MODEL.MASK_FORMER.TEST.PANOPTIC_ON or cfg.MODEL.MASK_FORMER.TEST.INSTANCE_ON ), "pixel_mean": cfg.MODEL.PIXEL_MEAN, "pixel_std": cfg.MODEL.PIXEL_STD, # inference "semantic_on": cfg.MODEL.MASK_FORMER.TEST.SEMANTIC_ON, "instance_on": cfg.MODEL.MASK_FORMER.TEST.INSTANCE_ON, "panoptic_on": cfg.MODEL.MASK_FORMER.TEST.PANOPTIC_ON, "test_topk_per_image": cfg.TEST.DETECTIONS_PER_IMAGE, "geometric_ensemble_alpha": cfg.MODEL.FC_CLIP.GEOMETRIC_ENSEMBLE_ALPHA, "geometric_ensemble_beta": cfg.MODEL.FC_CLIP.GEOMETRIC_ENSEMBLE_BETA, } @property def device(self): return self.pixel_mean.device def forward(self, batched_inputs): """ Args: batched_inputs: a list, batched outputs of :class:`DatasetMapper`. Each item in the list contains the inputs for one image. For now, each item in the list is a dict that contains: * "image": Tensor, image in (C, H, W) format. * "instances": per-region ground truth * Other information that's included in the original dicts, such as: "height", "width" (int): the output resolution of the model (may be different from input resolution), used in inference. Returns: list[dict]: each dict has the results for one image. The dict contains the following keys: * "sem_seg": A Tensor that represents the per-pixel segmentation prediced by the head. The prediction has shape KxHxW that represents the logits of each class for each pixel. * "panoptic_seg": A tuple that represent panoptic output panoptic_seg (Tensor): of shape (height, width) where the values are ids for each segment. segments_info (list[dict]): Describe each segment in `panoptic_seg`. Each dict contains keys "id", "category_id", "isthing". """ images = [x["image"].to(self.device) for x in batched_inputs] images = [(x - self.pixel_mean) / self.pixel_std for x in images] images = ImageList.from_tensors(images, self.size_divisibility) features = self.backbone(images.tensor) text_classifier, num_templates = self.get_text_classifier() # Append void class weight text_classifier = torch.cat([text_classifier, F.normalize(self.void_embedding.weight, dim=-1)], dim=0) features['text_classifier'] = text_classifier features['num_templates'] = num_templates outputs = self.sem_seg_head(features) if self.training: # mask classification target if "instances" in batched_inputs[0]: gt_instances = [x["instances"].to(self.device) for x in batched_inputs] targets = self.prepare_targets(gt_instances, images) else: targets = None # bipartite matching-based loss losses = self.criterion(outputs, targets) for k in list(losses.keys()): if k in self.criterion.weight_dict: losses[k] *= self.criterion.weight_dict[k] else: # remove this loss if not specified in `weight_dict` losses.pop(k) return losses else: mask_cls_results = outputs["pred_logits"] mask_pred_results = outputs["pred_masks"] # We ensemble the pred logits of in-vocab and out-vocab clip_feature = features["clip_vis_dense"] mask_for_pooling = F.interpolate(mask_pred_results, size=clip_feature.shape[-2:], mode='bilinear', align_corners=False) pooled_clip_feature = self.mask_pooling(clip_feature, mask_for_pooling) pooled_clip_feature = self.backbone.visual_prediction_forward(pooled_clip_feature) out_vocab_cls_results = get_classification_logits(pooled_clip_feature, text_classifier, self.backbone.clip_model.logit_scale, num_templates) in_vocab_cls_results = mask_cls_results[..., :-1] # remove void out_vocab_cls_results = out_vocab_cls_results[..., :-1] # remove void # Reference: https://github.com/NVlabs/ODISE/blob/main/odise/modeling/meta_arch/odise.py#L1506 out_vocab_cls_probs = out_vocab_cls_results.softmax(-1) in_vocab_cls_results = in_vocab_cls_results.softmax(-1) category_overlapping_mask = self.category_overlapping_mask.to(self.device) alpha = self.geometric_ensemble_alpha beta = self.geometric_ensemble_beta cls_logits_seen = ( (in_vocab_cls_results ** (1 - alpha) * out_vocab_cls_probs**alpha).log() * category_overlapping_mask ) cls_logits_unseen = ( (in_vocab_cls_results ** (1 - beta) * out_vocab_cls_probs**beta).log() * (1 - category_overlapping_mask) ) cls_results = cls_logits_seen + cls_logits_unseen # This is used to filtering void predictions. is_void_prob = F.softmax(mask_cls_results, dim=-1)[..., -1:] mask_cls_probs = torch.cat([ cls_results.softmax(-1) * (1.0 - is_void_prob), is_void_prob], dim=-1) mask_cls_results = torch.log(mask_cls_probs + 1e-8) # upsample masks mask_pred_results = F.interpolate( mask_pred_results, size=(images.tensor.shape[-2], images.tensor.shape[-1]), mode="bilinear", align_corners=False, ) del outputs processed_results = [] for mask_cls_result, mask_pred_result, input_per_image, image_size in zip( mask_cls_results, mask_pred_results, batched_inputs, images.image_sizes ): height = input_per_image.get("height", image_size[0]) width = input_per_image.get("width", image_size[1]) processed_results.append({}) if self.sem_seg_postprocess_before_inference: mask_pred_result = retry_if_cuda_oom(sem_seg_postprocess)( mask_pred_result, image_size, height, width ) mask_cls_result = mask_cls_result.to(mask_pred_result) # semantic segmentation inference if self.semantic_on: r = retry_if_cuda_oom(self.semantic_inference)(mask_cls_result, mask_pred_result) if not self.sem_seg_postprocess_before_inference: r = retry_if_cuda_oom(sem_seg_postprocess)(r, image_size, height, width) processed_results[-1]["sem_seg"] = r # panoptic segmentation inference if self.panoptic_on: panoptic_r = retry_if_cuda_oom(self.panoptic_inference)(mask_cls_result, mask_pred_result) processed_results[-1]["panoptic_seg"] = panoptic_r # instance segmentation inference if self.instance_on: instance_r = retry_if_cuda_oom(self.instance_inference)(mask_cls_result, mask_pred_result) processed_results[-1]["instances"] = instance_r return processed_results def prepare_targets(self, targets, images): h_pad, w_pad = images.tensor.shape[-2:] new_targets = [] for targets_per_image in targets: # pad gt gt_masks = targets_per_image.gt_masks padded_masks = torch.zeros((gt_masks.shape[0], h_pad, w_pad), dtype=gt_masks.dtype, device=gt_masks.device) padded_masks[:, : gt_masks.shape[1], : gt_masks.shape[2]] = gt_masks new_targets.append( { "labels": targets_per_image.gt_classes, "masks": padded_masks, } ) return new_targets def semantic_inference(self, mask_cls, mask_pred): mask_cls = F.softmax(mask_cls, dim=-1)[..., :-1] mask_pred = mask_pred.sigmoid() semseg = torch.einsum("qc,qhw->chw", mask_cls, mask_pred) return semseg def panoptic_inference(self, mask_cls, mask_pred): scores, labels = F.softmax(mask_cls, dim=-1).max(-1) mask_pred = mask_pred.sigmoid() num_classes = len(self.test_metadata.stuff_classes) keep = labels.ne(num_classes) & (scores > self.object_mask_threshold) cur_scores = scores[keep] cur_classes = labels[keep] cur_masks = mask_pred[keep] cur_mask_cls = mask_cls[keep] cur_mask_cls = cur_mask_cls[:, :-1] cur_prob_masks = cur_scores.view(-1, 1, 1) * cur_masks h, w = cur_masks.shape[-2:] panoptic_seg = torch.zeros((h, w), dtype=torch.int32, device=cur_masks.device) segments_info = [] current_segment_id = 0 if cur_masks.shape[0] == 0: # We didn't detect any mask :( return panoptic_seg, segments_info else: # take argmax cur_mask_ids = cur_prob_masks.argmax(0) stuff_memory_list = {} for k in range(cur_classes.shape[0]): pred_class = cur_classes[k].item() isthing = pred_class in self.test_metadata.thing_dataset_id_to_contiguous_id.values() mask_area = (cur_mask_ids == k).sum().item() original_area = (cur_masks[k] >= 0.5).sum().item() mask = (cur_mask_ids == k) & (cur_masks[k] >= 0.5) if mask_area > 0 and original_area > 0 and mask.sum().item() > 0: if mask_area / original_area < self.overlap_threshold: continue # merge stuff regions if not isthing: if int(pred_class) in stuff_memory_list.keys(): panoptic_seg[mask] = stuff_memory_list[int(pred_class)] continue else: stuff_memory_list[int(pred_class)] = current_segment_id + 1 current_segment_id += 1 panoptic_seg[mask] = current_segment_id segments_info.append( { "id": current_segment_id, "isthing": bool(isthing), "category_id": int(pred_class), } ) return panoptic_seg, segments_info def instance_inference(self, mask_cls, mask_pred): # mask_pred is already processed to have the same shape as original input image_size = mask_pred.shape[-2:] # [Q, K] scores = F.softmax(mask_cls, dim=-1)[:, :-1] # if this is panoptic segmentation if self.panoptic_on: num_classes = len(self.test_metadata.stuff_classes) else: num_classes = len(self.test_metadata.thing_classes) labels = torch.arange(num_classes, device=self.device).unsqueeze(0).repeat(self.num_queries, 1).flatten(0, 1) # scores_per_image, topk_indices = scores.flatten(0, 1).topk(self.num_queries, sorted=False) scores_per_image, topk_indices = scores.flatten(0, 1).topk(self.test_topk_per_image, sorted=False) labels_per_image = labels[topk_indices] topk_indices = topk_indices // num_classes # mask_pred = mask_pred.unsqueeze(1).repeat(1, self.sem_seg_head.num_classes, 1).flatten(0, 1) mask_pred = mask_pred[topk_indices] # if this is panoptic segmentation, we only keep the "thing" classes if self.panoptic_on: keep = torch.zeros_like(scores_per_image).bool() for i, lab in enumerate(labels_per_image): keep[i] = lab in self.test_metadata.thing_dataset_id_to_contiguous_id.values() scores_per_image = scores_per_image[keep] labels_per_image = labels_per_image[keep] mask_pred = mask_pred[keep] result = Instances(image_size) # mask (before sigmoid) result.pred_masks = (mask_pred > 0).float() result.pred_boxes = Boxes(torch.zeros(mask_pred.size(0), 4)) # Uncomment the following to get boxes from masks (this is slow) # result.pred_boxes = BitMasks(mask_pred > 0).get_bounding_boxes() # calculate average mask prob mask_scores_per_image = (mask_pred.sigmoid().flatten(1) * result.pred_masks.flatten(1)).sum(1) / (result.pred_masks.flatten(1).sum(1) + 1e-6) result.scores = scores_per_image * mask_scores_per_image result.pred_classes = labels_per_image return result